I wrote a client for FRED like fredpi except I used c#/dotnet. In the course of writing my client I came across some ambiguities in the FRED documentation. I also found some cases where the data I received from the API was not consistent with the documented API.

I wrote the FRED team several times and their responses only raised more questions. I eventually wound up creating my own definitions and writing my code against my own spec. I have long wondered, however, if I am not understanding the documentation correctly or if perhaps I am not using the API in the way it is intended to be used. After running a few queries with fredapi I see some cases where the same the same anomalies I found in the API are apparent in the the results returned by fredapi.

1.) Why does get latest data known on a given date return multiple rows for a given observation date? I would expect that since we are querying for latest we would only see one row e.g. the latest one. For the example shown here I would expect to see only these rows:

2013年10月01日 2014年03月27日 17089.6
2014年01月01日 2014年05月29日 17101.3

2.) Not all vintage dates are returned for a realtime query that spans multiple vintage periods:

fredapi documentation:

get_series_all_releases(self, series_id, realtime_start=None, realtime_end=None)
Get all data for a Fred series id including first releases and all revisions. This returns a DataFrame
with three columns: 'date', 'realtime_start', and 'value'. For instance, the US GDP for Q4 2013 was first released
to be 17102.5 on 2014年01月30日, and then revised to 17080.7 on 2014年02月28日, and then revised to 17089.6 on
2014年03月27日. You will therefore get three rows with the same 'date' (observation date) of 2013年10月01日 but three
different 'realtime_start' of 2014年01月30日, 2014年02月28日, and 2014年03月27日 with corresponding 'value' of 17102.5, 17080.7
and 17089.6

This query:

fred.get_series('GDP', realtime_start='2014-01-28', realtime_end='2014-03-30', observation_start='2013-10-01', observation_end='2013-10-01')

returns only one result:

2013年10月01日 17089.6

Based on the spreadsheet for GDP downloaded from Alfred, three vintages for observation period 2013年10月01日 were released between 2014年01月28日 and 2014年03月30日:

2014年01月30日	17102.5
2014年02月28日	17080.7
2014年03月27日	17089.6

Again note that the fredapi documentation says "...including first releases and all revisions" therefore I would expect to see all three vintages. Also, note that the date returned by the query is not a vintage date but instead appears to be the observation period.

Is the result of this query correct? Based on FRED documentation, result from the query above demonstrates what I believe is a defect in the FRED API (not a defect in fredapi). I believe FRED has co-mingled the concepts of realtime periods and vintage dates. As a result, queries such as the one above return meaningless or inconsistent results. I posted a fairly in-depth analysis of my findings I will provide a link if requested.
Capture